Page MenuHomeFreeBSD
Differential D24232 Diff 70031 sys/net/route/nhop_ctl.c

Changeset View

Standalone View

View Options

sys/net/route/nhop_ctl.c

  • This file was added.
/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (c) 2020 Alexander V. Chernikov
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* $FreeBSD$
*/
#define RTDEBUG
#include <sys/cdefs.h>
#include "opt_inet.h"
#include "opt_route.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/lock.h>
#include <sys/rwlock.h>
#include <sys/malloc.h>
#include <sys/socket.h>
#include <sys/sysctl.h>
#include <sys/kernel.h>
#include <sys/epoch.h>
#include <net/if.h>
#include <net/if_var.h>
#include <net/if_dl.h>
#include <net/route.h>
#include <net/route_var.h>
#include <net/vnet.h>
#include <netinet/in.h>
#include <netinet/in_var.h>
#include <netinet/in_fib.h>
#include <net/route/nhop_utils.h>
#include <net/route/nhop.h>
#include <net/route/nhop_var.h>
#include <net/route/shared.h>
/*
* This file contains core functionality for the nexthop ("nhop") route subsystem.
* The business logic needed to create nexhop objects is implemented here.
*
* Nexthops in the original sense are the objects containing all the necessary
* information to forward the packet to the selected destination.
* In particular, nexthop is defined by a combination of
* ifp, ifa, aifp, mtu, gw addr(if set), nh_type, nh_family, mask of rt_flags and
* NHF_DEFAULT
*
* Additionally, each nexthop gets assigned its unique index (nexthop index).
* It serves two purposes: first one is to ease the ability of userland programs to
* reference nexthops by their index. The second one allows lookup algorithms to
* to store index instead of pointer (2 bytes vs 8) as a lookup result.
* All nexthops are stored in the resizable hash table.
*
* Basically, this file revolves around supporting 2 functions:
* 1) fill_nhop(), which contains all business logic on filling the nexthop fields
* based on the provided request
* 2) nhop_get(), which gets a nexthop based on the provided request.
*
* Conventions:
* 1) non-exported functions start with verb
* 2) exported function starts with the subsystem prefix: "nhop"
*
*/
static int fill_nhop(const struct nhop_request *req, struct nhop_object *nh);
static int dump_nhop_entry(struct rib_head *rh, struct nhop_object *nh, struct sysctl_req *w);
static struct nhop_priv *alloc_nhop_partial(const struct nhop_request *req);
static int finalize_nhop(struct nhop_priv *nh_priv, const struct nhop_request *req);
static struct ifnet *get_aifp(const struct nhop_request *req, int reference);
static void fill_sdl_from_ifp(struct sockaddr_dl_short *sdl, const struct ifnet *ifp);
static void destroy_nhop_epoch(epoch_context_t ctx);
static void destroy_nhop(struct nhop_priv *nh_priv);
static void print_nhop(const char *prefix, const struct nhop_object *nh);
_Static_assert(__offsetof(struct nhop_object, nh_ifp) == 32,
"nhop_object: wrong nh_ifp offset");
_Static_assert(sizeof(struct nhop_object) <= 128,
"nhop_object: size exceeds 128 bytes");
static uma_zone_t nhops_zone; /* Global zone for each and every nexthop */
#define NHOP_OBJECT_ALIGNED_SIZE roundup2(sizeof(struct nhop_object), \
2 * CACHE_LINE_SIZE)
#define NHOP_PRIV_ALIGNED_SIZE roundup2(sizeof(struct nhop_priv), \
2 * CACHE_LINE_SIZE)
void
nhops_init(void)
{
nhops_zone = uma_zcreate("routing nhops",
NHOP_OBJECT_ALIGNED_SIZE + NHOP_PRIV_ALIGNED_SIZE,
NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
}
/*
* Fetches the interface of source address used by the route.
* In all cases except interface-address-route it would be the
* same as the transmit interfaces.
* However, for the interface address this function will return
* this interface ifp instead of loopback. This is needed to support
* link-local IPv6 loopback communications.
*
* If @reference is non-zero, found ifp is referenced.
*
* Returns found ifp.
*/
static struct ifnet *
get_aifp(const struct nhop_request *req, int reference)
{
struct ifnet *aifp = NULL;
struct sockaddr_dl *sdl;
struct epoch_tracker et;
/*
* Adjust the "outgoing" interface. If we're going to loop
* the packet back to ourselves, the ifp would be the loopback
* interface. However, we'd rather know the interface associated
* to the destination address (which should probably be one of
* our own addresses.)
*/
if ((req->ifp->if_flags & IFF_LOOPBACK) &&
req->gw->sa_family == AF_LINK) {
sdl = (struct sockaddr_dl *)req->gw;
NET_EPOCH_ENTER(et);
if (reference)
aifp = ifnet_byindex_ref(sdl->sdl_index);
else
aifp = ifnet_byindex(sdl->sdl_index);
NET_EPOCH_EXIT(et);
if (aifp == NULL) {
DPRINTF("unable to get aifp for %s index %d",
if_name(req->ifp), sdl->sdl_index);
}
}
if (aifp == NULL) {
aifp = req->ifp;
if (reference)
if_ref(aifp);
}
return (aifp);
}
int
cmp_priv(const struct nhop_priv *_one, const struct nhop_priv *_two)
{
if (memcmp(_one->nh, _two->nh, NHOP_END_CMP) != 0)
return (0);
if ((_one->nh_type != _two->nh_type) ||
(_one->nh_family != _two->nh_family))
return (0);
return (1);
}
/*
* Finds or creates new nhop_object based on @req.
* Returns referenced and linked nhop_object or NULL.
*/
__noinline struct nhop_object *
nhop_get(struct rib_head *rh, const struct nhop_request *req)
{
struct nh_control *ctl = rh->nh_control;
struct nhop_priv *nh_priv, *nh_tmp_priv;
/*
* In the cases with large amount of routes, most routes will
* share nexthops, making alloc-to-lookup ratio low.
*
* With that in mind, split nexthop allocation in two stages:
* first does the bare minimum to support the lookup for the
* existing nexthop.
* the second does the heavy-lifting with counters allocations
* and external objects refcounting.
*/
nh_tmp_priv = alloc_nhop_partial(req);
if (nh_tmp_priv == NULL) {
RTSTAT_INC(rts_nh_alloc_failure);
DPRINTF("nh_alloc failed");
return (NULL);
}
nh_priv = find_nhop(ctl, nh_tmp_priv);
if (nh_priv != NULL) {
uma_zfree(nhops_zone, nh_tmp_priv->nh);
return (nh_priv->nh);
}
nh_priv = nh_tmp_priv;
if (finalize_nhop(nh_priv, req) != 0) {
RTSTAT_INC(rts_nh_alloc_failure);
DPRINTF("nh_alloc_finalize failed");
return (NULL);
}
if (link_nhop(ctl, nh_priv) == 0) {
/*
* Adding nexthop to the datastructures
* failed. Call destructor w/o waiting for
* the epoch end, as nexthop is not used
* and return.
*/
DPRINTF("link_nhop failed!");
destroy_nhop(nh_priv);
return (NULL);
}
return (nh_priv->nh);
}
/*
* Fills in shorted link-level sockadd version suitable to be stored inside the
* nexthop gateway buffer.
*/
static void
fill_sdl_from_ifp(struct sockaddr_dl_short *sdl, const struct ifnet *ifp)
{
sdl->sdl_family = AF_LINK;
sdl->sdl_len = sizeof(struct sockaddr_dl_short);
sdl->sdl_index = ifp->if_index;
sdl->sdl_type = ifp->if_type;
}
/*
* Allocates nexthops and fills in the minimum amount of
* data to perform hash lookup.
* No external objects are referenced. Packet counters
* not allocated.
*
* Returns nh_priv pointer or NULL.
*/
static struct nhop_priv *
alloc_nhop_partial(const struct nhop_request *req)
{
struct nhop_object *nh;
struct nhop_priv *nh_priv;
KASSERT((req->mtu > 0), ("nh requested mtu is zero"));
nh = (struct nhop_object *)uma_zalloc(nhops_zone, M_NOWAIT | M_ZERO);
if (nh == NULL)
return (NULL);
nh_priv = (struct nhop_priv *)((char *)nh + NHOP_OBJECT_ALIGNED_SIZE);
nh->nh_priv = nh_priv;
nh_priv->nh = nh;
if (fill_nhop(req, nh) != 0) {
uma_zfree(nhops_zone, nh);
return (NULL);
}
/* calculate aifp, but don't reference it */
nh->nh_aifp = get_aifp(req, 0);
return (nh_priv);
}
/*
* Finalizes nexthop data to make nhop suitable for linking into the
* hash table.
* Returns 0 if successful,
* errno otherwise. @nh_priv is freed in case of error.
*/
static int
finalize_nhop(struct nhop_priv *nh_priv, const struct nhop_request *req)
{
struct nhop_object *nh;
nh = nh_priv->nh;
/* Allocate per-cpu packet counter */
nh->nh_pksent = counter_u64_alloc(M_NOWAIT);
if (nh->nh_pksent == NULL) {
uma_zfree(nhops_zone, nh);
return (ENOMEM);
}
/* Reference external objects and calculate (referenced) ifa */
if_ref(nh->nh_ifp);
ifa_ref(nh->nh_ifa);
nh->nh_aifp = get_aifp(req, 1);
DPRINTF("AIFP: %p req->ifp %p nh_ifp %p", nh->nh_aifp, req->ifp, nh->nh_ifp);
NH_PRIV_LOCK_INIT(nh_priv);
refcount_init(&nh_priv->nh_refcnt, 1);
print_nhop("FINALIZE", nh);
return (0);
}
static void
print_nhop_sa(char *buf, size_t buflen, const struct sockaddr *sa)
{
if (sa->sa_family == AF_INET) {
const struct sockaddr_in *sin4;
sin4 = (const struct sockaddr_in *)sa;
inet_ntop(AF_INET, &sin4->sin_addr, buf, buflen);
} else if (sa->sa_family == AF_INET6) {
const struct sockaddr_in6 *sin6;
sin6 = (const struct sockaddr_in6 *)sa;
inet_ntop(AF_INET6, &sin6->sin6_addr, buf, buflen);
} else if (sa->sa_family == AF_LINK) {
const struct sockaddr_dl *sdl;
sdl = (const struct sockaddr_dl *)sa;
snprintf(buf, buflen, "if#%d", sdl->sdl_index);
} else
snprintf(buf, buflen, "af:%d", sa->sa_family);
}
static void
print_nhop(const char *prefix, const struct nhop_object *nh)
{
char src_buf[INET6_ADDRSTRLEN], addr_buf[INET6_ADDRSTRLEN];
int af = nh->nh_priv->nh_family;
print_nhop_sa(src_buf, sizeof(src_buf), nh->nh_ifa->ifa_addr);
print_nhop_sa(addr_buf, sizeof(addr_buf), &nh->gw_sa);
DPRINTF("%s nhop priv %p: AF %d ifp %p %s addr %s src %p %s aifp %p %s mtu %d nh_flags %X",
prefix, nh->nh_priv, af, nh->nh_ifp, if_name(nh->nh_ifp), addr_buf,
nh->nh_ifa, src_buf, nh->nh_aifp, if_name(nh->nh_aifp), nh->nh_mtu,
nh->nh_flags);
}
static void
destroy_nhop(struct nhop_priv *nh_priv)
{
aeUnsubmitted
Done
Inline Actions

this lock also seems unneeded.

ae: this lock also seems unneeded.
struct nhop_object *nh;
nh = nh_priv->nh;
NH_PRIV_LOCK(nh_priv);
print_nhop("DEL", nh);
NH_PRIV_UNLOCK(nh_priv);
if_rele(nh->nh_ifp);
if_rele(nh->nh_aifp);
ifa_free(nh->nh_ifa);
counter_u64_free(nh->nh_pksent);
uma_zfree(nhops_zone, nh);
}
/*
* Epoch callback indicating nhop is safe to destroy
*/
static void
destroy_nhop_epoch(epoch_context_t ctx)
{
struct nhop_priv *nh_priv;
nh_priv = __containerof(ctx, struct nhop_priv, nh_epoch_ctx);
destroy_nhop(nh_priv);
}
/*
* Fills @nh fields with the data supplied in the @req.
* Function does NOT fill in nh_aifp and does not take any reference.
* Returns 0 on success.
*/
static int
fill_nhop(const struct nhop_request *req, struct nhop_object *nh)
{
int rt_flags;
rt_flags = req->rt_flags & NHOP_RT_FLAG_MASK;
nh->nh_ifp = req->ifp;
nh->nh_mtu = req->mtu;
nh->nh_flags = fib_rte_to_nh_flags(rt_flags);
nh->nh_flags |= (req->nh_flags_additional & NHF_DEFAULT);
nh->nh_priv->rt_flags = rt_flags;
nh->nh_ifa = req->ifa;
if (req->rt_flags & RTF_GATEWAY) {
if (req->gw->sa_len > sizeof(struct sockaddr_in6)) {
DPRINTF("nhop SA size too big: AF %d len %u",
req->gw->sa_family, req->gw->sa_len);
return (ENOMEM);
}
memcpy(&nh->gw_sa, req->gw, req->gw->sa_len);
} else {
/*
* Interface route. Currently the route.c code adds
* empty sa of type AF_LINK, which is 56 bytes long.
* The only place where this data is used is the IPv6
* loopback output, where we need to preserve the original
* interface to maintain proper scoping.
* Current code stores original interface in the separate field
* (nh_aifp, see below). Given that, write fake empty SA
* with the request AF.
*/
fill_sdl_from_ifp(&nh->gwl_sa, req->ifp);
}
nh->nh_priv->nh_family = req->family;
nh->nh_priv->nh_type = req->nh_type;
return (0);
}
int
nhop_ref_object(struct nhop_object *nh)
{
return (refcount_acquire_if_not_zero(&nh->nh_priv->nh_refcnt));
}
void
nhop_free_object(struct nhop_object *nh)
{
struct nh_control *ctl;
struct nhop_priv *nh_priv = nh->nh_priv;
if (!refcount_release(&nh_priv->nh_refcnt))
return;
NH_PRIV_LOCK(nh_priv);
ctl = nh_priv->nh_control;
/* Use nh_control as an indicator of linked/unlinked entry */
nh_priv->nh_control = NULL;
NH_PRIV_UNLOCK(nh_priv);
if (ctl != NULL) {
if (unlink_nhop(ctl, nh_priv) == NULL) {
/* Do not try to reclaim */
DPRINTF("Failed to find nexhop %p", nh_priv);
return;
}
}
epoch_call(net_epoch_preempt, destroy_nhop_epoch,
&nh_priv->nh_epoch_ctx);
}
int
nhop_ref_any(struct nhop_object *nh)
{
return (nhop_ref_object(nh));
}
void
nhop_free_any(struct nhop_object *nh)
{
nhop_free_object(nh);
}
/* Helper functions */
uint32_t
nhop_get_idx(const struct nhop_object *nh)
{
return (nh->nh_priv->nh_idx);
}
void
nhops_update_ifmtu(struct rib_head *rh, struct ifnet *ifp, uint32_t mtu)
{
struct nh_control *ctl;
struct nhop_priv *nh_priv;
struct nhop_object *nh;
ctl = rh->nh_control;
NHOPS_RLOCK(ctl);
CHT_SLIST_FOREACH(&ctl->nh_head, nhops, nh_priv) {
nh = nh_priv->nh;
if (nh->nh_ifp == ifp) {
if ((nh_priv->rt_flags & RTF_FIXEDMTU) == 0 ||
nh->nh_mtu > mtu) {
/* Update */
NH_PRIV_LOCK(nh_priv);
nh->nh_mtu = mtu;
NH_PRIV_UNLOCK(nh_priv);
}
}
} CHT_SLIST_FOREACH_END;
NHOPS_RUNLOCK(ctl);
}
/*
* Dumps a single entry to sysctl buffer.
*
* Layout:
* rt_msghdr - generic RTM header to allow users to skip non-understood messages
* nhop_external - nexhop description structure (with length)
* nhop_addrs - structure encapsulating GW/SRC sockaddrs
*/
static int
dump_nhop_entry(struct rib_head *rh, struct nhop_object *nh, struct sysctl_req *w)
{
struct {
struct rt_msghdr rtm;
struct nhop_external nhe;
struct nhop_addrs na;
} arpc;
struct nhop_external *pnhe;
struct sockaddr *gw_sa, *src_sa;
struct sockaddr_storage ss;
size_t addrs_len;
int error;
//DPRINTF("Dumping: head %p nh %p flags %X req %p\n", rh, nh, nh->nh_flags, w);
memset(&arpc, 0, sizeof(arpc));
arpc.rtm.rtm_msglen = sizeof(arpc);
arpc.rtm.rtm_version = RTM_VERSION;
arpc.rtm.rtm_type = RTM_GET;
//arpc.rtm.rtm_flags = RTF_UP;
arpc.rtm.rtm_flags = nh->nh_priv->rt_flags;
/* nhop_external */
pnhe = &arpc.nhe;
pnhe->nh_len = sizeof(struct nhop_external);
pnhe->nh_idx = nh->nh_priv->nh_idx;
pnhe->nh_fib = rh->rib_fibnum;
pnhe->ifindex = nh->nh_ifp->if_index;
pnhe->aifindex = nh->nh_aifp->if_index;
pnhe->nh_family = nh->nh_priv->nh_family;
pnhe->nh_type = nh->nh_priv->nh_type;
pnhe->nh_mtu = nh->nh_mtu;
pnhe->nh_flags = nh->nh_flags;
memcpy(pnhe->nh_prepend, nh->nh_prepend, sizeof(nh->nh_prepend));
pnhe->prepend_len = nh->nh_prepend_len;
pnhe->nh_refcount = nh->nh_priv->nh_refcnt;
pnhe->nh_pksent = counter_u64_fetch(nh->nh_pksent);
/* sockaddr container */
addrs_len = sizeof(struct nhop_addrs);
arpc.na.gw_sa_off = addrs_len;
gw_sa = (struct sockaddr *)&nh->gw4_sa;
addrs_len += gw_sa->sa_len;
src_sa = nh->nh_ifa->ifa_addr;
if (src_sa->sa_family == AF_LINK) {
/* Shorten structure */
memset(&ss, 0, sizeof(struct sockaddr_storage));
fill_sdl_from_ifp((struct sockaddr_dl_short *)&ss,
nh->nh_ifa->ifa_ifp);
src_sa = (struct sockaddr *)&ss;
}
arpc.na.src_sa_off = addrs_len;
addrs_len += src_sa->sa_len;
/* Write total length */
arpc.na.na_len = addrs_len;
arpc.rtm.rtm_msglen += arpc.na.na_len - sizeof(struct nhop_addrs);
error = SYSCTL_OUT(w, &arpc, sizeof(arpc));
if (error == 0)
error = SYSCTL_OUT(w, gw_sa, gw_sa->sa_len);
if (error == 0)
error = SYSCTL_OUT(w, src_sa, src_sa->sa_len);
/*
DPRINTF("Exported %d ifindex %d family %d type %d error %d\n", nh->nh_priv->nh_idx, pnhe->ifindex,
pnhe->nh_family, pnhe->nh_type, error);
*/
return (error);
}
int
nhops_dump_sysctl(struct rib_head *rh, struct sysctl_req *w)
{
struct nh_control *ctl;
struct nhop_priv *nh_priv;
int error;
ctl = rh->nh_control;
NHOPS_RLOCK(ctl);
DPRINTF("NHDUMP: count=%u", ctl->nh_head.items_count);
CHT_SLIST_FOREACH(&ctl->nh_head, nhops, nh_priv) {
error = dump_nhop_entry(rh, nh_priv->nh, w);
if (error != 0)
return (error);
} CHT_SLIST_FOREACH_END;
NHOPS_RUNLOCK(ctl);
return (0);
}